mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / fat / inode.c
blobc41393e30a04a26ab87e2422299b65d0b41f9de4
1 /*
2 * linux/fs/fat/inode.c
4 * Written 1992,1993 by Werner Almesberger
5 * VFAT extensions by Gordon Chaffee, merged with msdos fs by Henrik Storner
6 * Rewritten for the constant inumbers support by Al Viro
8 * Fixes:
10 * Max Cohan: Fixed invalid FSINFO offset when info_sector is 0
13 #include <linux/module.h>
14 #include <linux/pagemap.h>
15 #include <linux/mpage.h>
16 #include <linux/vfs.h>
17 #include <linux/seq_file.h>
18 #include <linux/parser.h>
19 #include <linux/uio.h>
20 #include <linux/blkdev.h>
21 #include <linux/backing-dev.h>
22 #include <asm/unaligned.h>
23 #include "fat.h"
25 #ifndef CONFIG_FAT_DEFAULT_IOCHARSET
26 /* if user don't select VFAT, this is undefined. */
27 #define CONFIG_FAT_DEFAULT_IOCHARSET ""
28 #endif
30 #define KB_IN_SECTORS 2
33 * A deserialized copy of the on-disk structure laid out in struct
34 * fat_boot_sector.
36 struct fat_bios_param_block {
37 u16 fat_sector_size;
38 u8 fat_sec_per_clus;
39 u16 fat_reserved;
40 u8 fat_fats;
41 u16 fat_dir_entries;
42 u16 fat_sectors;
43 u16 fat_fat_length;
44 u32 fat_total_sect;
46 u8 fat16_state;
47 u32 fat16_vol_id;
49 u32 fat32_length;
50 u32 fat32_root_cluster;
51 u16 fat32_info_sector;
52 u8 fat32_state;
53 u32 fat32_vol_id;
56 static int fat_default_codepage = CONFIG_FAT_DEFAULT_CODEPAGE;
57 static char fat_default_iocharset[] = CONFIG_FAT_DEFAULT_IOCHARSET;
59 static struct fat_floppy_defaults {
60 unsigned nr_sectors;
61 unsigned sec_per_clus;
62 unsigned dir_entries;
63 unsigned media;
64 unsigned fat_length;
65 } floppy_defaults[] = {
67 .nr_sectors = 160 * KB_IN_SECTORS,
68 .sec_per_clus = 1,
69 .dir_entries = 64,
70 .media = 0xFE,
71 .fat_length = 1,
74 .nr_sectors = 180 * KB_IN_SECTORS,
75 .sec_per_clus = 1,
76 .dir_entries = 64,
77 .media = 0xFC,
78 .fat_length = 2,
81 .nr_sectors = 320 * KB_IN_SECTORS,
82 .sec_per_clus = 2,
83 .dir_entries = 112,
84 .media = 0xFF,
85 .fat_length = 1,
88 .nr_sectors = 360 * KB_IN_SECTORS,
89 .sec_per_clus = 2,
90 .dir_entries = 112,
91 .media = 0xFD,
92 .fat_length = 2,
96 int fat_add_cluster(struct inode *inode)
98 int err, cluster;
100 err = fat_alloc_clusters(inode, &cluster, 1);
101 if (err)
102 return err;
103 /* FIXME: this cluster should be added after data of this
104 * cluster is writed */
105 err = fat_chain_add(inode, cluster, 1);
106 if (err)
107 fat_free_clusters(inode, cluster);
108 return err;
111 static inline int __fat_get_block(struct inode *inode, sector_t iblock,
112 unsigned long *max_blocks,
113 struct buffer_head *bh_result, int create)
115 struct super_block *sb = inode->i_sb;
116 struct msdos_sb_info *sbi = MSDOS_SB(sb);
117 unsigned long mapped_blocks;
118 sector_t phys, last_block;
119 int err, offset;
121 err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create, false);
122 if (err)
123 return err;
124 if (phys) {
125 map_bh(bh_result, sb, phys);
126 *max_blocks = min(mapped_blocks, *max_blocks);
127 return 0;
129 if (!create)
130 return 0;
132 if (iblock != MSDOS_I(inode)->mmu_private >> sb->s_blocksize_bits) {
133 fat_fs_error(sb, "corrupted file size (i_pos %lld, %lld)",
134 MSDOS_I(inode)->i_pos, MSDOS_I(inode)->mmu_private);
135 return -EIO;
138 last_block = inode->i_blocks >> (sb->s_blocksize_bits - 9);
139 offset = (unsigned long)iblock & (sbi->sec_per_clus - 1);
141 * allocate a cluster according to the following.
142 * 1) no more available blocks
143 * 2) not part of fallocate region
145 if (!offset && !(iblock < last_block)) {
146 /* TODO: multiple cluster allocation would be desirable. */
147 err = fat_add_cluster(inode);
148 if (err)
149 return err;
151 /* available blocks on this cluster */
152 mapped_blocks = sbi->sec_per_clus - offset;
154 *max_blocks = min(mapped_blocks, *max_blocks);
155 MSDOS_I(inode)->mmu_private += *max_blocks << sb->s_blocksize_bits;
157 err = fat_bmap(inode, iblock, &phys, &mapped_blocks, create, false);
158 if (err)
159 return err;
161 BUG_ON(!phys);
162 BUG_ON(*max_blocks != mapped_blocks);
163 set_buffer_new(bh_result);
164 map_bh(bh_result, sb, phys);
166 return 0;
169 static int fat_get_block(struct inode *inode, sector_t iblock,
170 struct buffer_head *bh_result, int create)
172 struct super_block *sb = inode->i_sb;
173 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
174 int err;
176 err = __fat_get_block(inode, iblock, &max_blocks, bh_result, create);
177 if (err)
178 return err;
179 bh_result->b_size = max_blocks << sb->s_blocksize_bits;
180 return 0;
183 static int fat_writepage(struct page *page, struct writeback_control *wbc)
185 return block_write_full_page(page, fat_get_block, wbc);
188 static int fat_writepages(struct address_space *mapping,
189 struct writeback_control *wbc)
191 return mpage_writepages(mapping, wbc, fat_get_block);
194 static int fat_readpage(struct file *file, struct page *page)
196 return mpage_readpage(page, fat_get_block);
199 static int fat_readpages(struct file *file, struct address_space *mapping,
200 struct list_head *pages, unsigned nr_pages)
202 return mpage_readpages(mapping, pages, nr_pages, fat_get_block);
205 static void fat_write_failed(struct address_space *mapping, loff_t to)
207 struct inode *inode = mapping->host;
209 if (to > inode->i_size) {
210 truncate_pagecache(inode, inode->i_size);
211 fat_truncate_blocks(inode, inode->i_size);
215 static int fat_write_begin(struct file *file, struct address_space *mapping,
216 loff_t pos, unsigned len, unsigned flags,
217 struct page **pagep, void **fsdata)
219 int err;
221 *pagep = NULL;
222 err = cont_write_begin(file, mapping, pos, len, flags,
223 pagep, fsdata, fat_get_block,
224 &MSDOS_I(mapping->host)->mmu_private);
225 if (err < 0)
226 fat_write_failed(mapping, pos + len);
227 return err;
230 static int fat_write_end(struct file *file, struct address_space *mapping,
231 loff_t pos, unsigned len, unsigned copied,
232 struct page *pagep, void *fsdata)
234 struct inode *inode = mapping->host;
235 int err;
236 err = generic_write_end(file, mapping, pos, len, copied, pagep, fsdata);
237 if (err < len)
238 fat_write_failed(mapping, pos + len);
239 if (!(err < 0) && !(MSDOS_I(inode)->i_attrs & ATTR_ARCH)) {
240 inode->i_mtime = inode->i_ctime = current_time(inode);
241 MSDOS_I(inode)->i_attrs |= ATTR_ARCH;
242 mark_inode_dirty(inode);
244 return err;
247 static ssize_t fat_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
249 struct file *file = iocb->ki_filp;
250 struct address_space *mapping = file->f_mapping;
251 struct inode *inode = mapping->host;
252 size_t count = iov_iter_count(iter);
253 loff_t offset = iocb->ki_pos;
254 ssize_t ret;
256 if (iov_iter_rw(iter) == WRITE) {
258 * FIXME: blockdev_direct_IO() doesn't use ->write_begin(),
259 * so we need to update the ->mmu_private to block boundary.
261 * But we must fill the remaining area or hole by nul for
262 * updating ->mmu_private.
264 * Return 0, and fallback to normal buffered write.
266 loff_t size = offset + count;
267 if (MSDOS_I(inode)->mmu_private < size)
268 return 0;
272 * FAT need to use the DIO_LOCKING for avoiding the race
273 * condition of fat_get_block() and ->truncate().
275 ret = blockdev_direct_IO(iocb, inode, iter, fat_get_block);
276 if (ret < 0 && iov_iter_rw(iter) == WRITE)
277 fat_write_failed(mapping, offset + count);
279 return ret;
282 static int fat_get_block_bmap(struct inode *inode, sector_t iblock,
283 struct buffer_head *bh_result, int create)
285 struct super_block *sb = inode->i_sb;
286 unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
287 int err;
288 sector_t bmap;
289 unsigned long mapped_blocks;
291 BUG_ON(create != 0);
293 err = fat_bmap(inode, iblock, &bmap, &mapped_blocks, create, true);
294 if (err)
295 return err;
297 if (bmap) {
298 map_bh(bh_result, sb, bmap);
299 max_blocks = min(mapped_blocks, max_blocks);
302 bh_result->b_size = max_blocks << sb->s_blocksize_bits;
304 return 0;
307 static sector_t _fat_bmap(struct address_space *mapping, sector_t block)
309 sector_t blocknr;
311 /* fat_get_cluster() assumes the requested blocknr isn't truncated. */
312 down_read(&MSDOS_I(mapping->host)->truncate_lock);
313 blocknr = generic_block_bmap(mapping, block, fat_get_block_bmap);
314 up_read(&MSDOS_I(mapping->host)->truncate_lock);
316 return blocknr;
320 * fat_block_truncate_page() zeroes out a mapping from file offset `from'
321 * up to the end of the block which corresponds to `from'.
322 * This is required during truncate to physically zeroout the tail end
323 * of that block so it doesn't yield old data if the file is later grown.
324 * Also, avoid causing failure from fsx for cases of "data past EOF"
326 int fat_block_truncate_page(struct inode *inode, loff_t from)
328 return block_truncate_page(inode->i_mapping, from, fat_get_block);
331 static const struct address_space_operations fat_aops = {
332 .readpage = fat_readpage,
333 .readpages = fat_readpages,
334 .writepage = fat_writepage,
335 .writepages = fat_writepages,
336 .write_begin = fat_write_begin,
337 .write_end = fat_write_end,
338 .direct_IO = fat_direct_IO,
339 .bmap = _fat_bmap
343 * New FAT inode stuff. We do the following:
344 * a) i_ino is constant and has nothing with on-disk location.
345 * b) FAT manages its own cache of directory entries.
346 * c) *This* cache is indexed by on-disk location.
347 * d) inode has an associated directory entry, all right, but
348 * it may be unhashed.
349 * e) currently entries are stored within struct inode. That should
350 * change.
351 * f) we deal with races in the following way:
352 * 1. readdir() and lookup() do FAT-dir-cache lookup.
353 * 2. rename() unhashes the F-d-c entry and rehashes it in
354 * a new place.
355 * 3. unlink() and rmdir() unhash F-d-c entry.
356 * 4. fat_write_inode() checks whether the thing is unhashed.
357 * If it is we silently return. If it isn't we do bread(),
358 * check if the location is still valid and retry if it
359 * isn't. Otherwise we do changes.
360 * 5. Spinlock is used to protect hash/unhash/location check/lookup
361 * 6. fat_evict_inode() unhashes the F-d-c entry.
362 * 7. lookup() and readdir() do igrab() if they find a F-d-c entry
363 * and consider negative result as cache miss.
366 static void fat_hash_init(struct super_block *sb)
368 struct msdos_sb_info *sbi = MSDOS_SB(sb);
369 int i;
371 spin_lock_init(&sbi->inode_hash_lock);
372 for (i = 0; i < FAT_HASH_SIZE; i++)
373 INIT_HLIST_HEAD(&sbi->inode_hashtable[i]);
376 static inline unsigned long fat_hash(loff_t i_pos)
378 return hash_32(i_pos, FAT_HASH_BITS);
381 static void dir_hash_init(struct super_block *sb)
383 struct msdos_sb_info *sbi = MSDOS_SB(sb);
384 int i;
386 spin_lock_init(&sbi->dir_hash_lock);
387 for (i = 0; i < FAT_HASH_SIZE; i++)
388 INIT_HLIST_HEAD(&sbi->dir_hashtable[i]);
391 void fat_attach(struct inode *inode, loff_t i_pos)
393 struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
395 if (inode->i_ino != MSDOS_ROOT_INO) {
396 struct hlist_head *head = sbi->inode_hashtable
397 + fat_hash(i_pos);
399 spin_lock(&sbi->inode_hash_lock);
400 MSDOS_I(inode)->i_pos = i_pos;
401 hlist_add_head(&MSDOS_I(inode)->i_fat_hash, head);
402 spin_unlock(&sbi->inode_hash_lock);
405 /* If NFS support is enabled, cache the mapping of start cluster
406 * to directory inode. This is used during reconnection of
407 * dentries to the filesystem root.
409 if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
410 struct hlist_head *d_head = sbi->dir_hashtable;
411 d_head += fat_dir_hash(MSDOS_I(inode)->i_logstart);
413 spin_lock(&sbi->dir_hash_lock);
414 hlist_add_head(&MSDOS_I(inode)->i_dir_hash, d_head);
415 spin_unlock(&sbi->dir_hash_lock);
418 EXPORT_SYMBOL_GPL(fat_attach);
420 void fat_detach(struct inode *inode)
422 struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
423 spin_lock(&sbi->inode_hash_lock);
424 MSDOS_I(inode)->i_pos = 0;
425 hlist_del_init(&MSDOS_I(inode)->i_fat_hash);
426 spin_unlock(&sbi->inode_hash_lock);
428 if (S_ISDIR(inode->i_mode) && sbi->options.nfs) {
429 spin_lock(&sbi->dir_hash_lock);
430 hlist_del_init(&MSDOS_I(inode)->i_dir_hash);
431 spin_unlock(&sbi->dir_hash_lock);
434 EXPORT_SYMBOL_GPL(fat_detach);
436 struct inode *fat_iget(struct super_block *sb, loff_t i_pos)
438 struct msdos_sb_info *sbi = MSDOS_SB(sb);
439 struct hlist_head *head = sbi->inode_hashtable + fat_hash(i_pos);
440 struct msdos_inode_info *i;
441 struct inode *inode = NULL;
443 spin_lock(&sbi->inode_hash_lock);
444 hlist_for_each_entry(i, head, i_fat_hash) {
445 BUG_ON(i->vfs_inode.i_sb != sb);
446 if (i->i_pos != i_pos)
447 continue;
448 inode = igrab(&i->vfs_inode);
449 if (inode)
450 break;
452 spin_unlock(&sbi->inode_hash_lock);
453 return inode;
456 static int is_exec(unsigned char *extension)
458 unsigned char exe_extensions[] = "EXECOMBAT", *walk;
460 for (walk = exe_extensions; *walk; walk += 3)
461 if (!strncmp(extension, walk, 3))
462 return 1;
463 return 0;
466 static int fat_calc_dir_size(struct inode *inode)
468 struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
469 int ret, fclus, dclus;
471 inode->i_size = 0;
472 if (MSDOS_I(inode)->i_start == 0)
473 return 0;
475 ret = fat_get_cluster(inode, FAT_ENT_EOF, &fclus, &dclus);
476 if (ret < 0)
477 return ret;
478 inode->i_size = (fclus + 1) << sbi->cluster_bits;
480 return 0;
483 static int fat_validate_dir(struct inode *dir)
485 struct super_block *sb = dir->i_sb;
487 if (dir->i_nlink < 2) {
488 /* Directory should have "."/".." entries at least. */
489 fat_fs_error(sb, "corrupted directory (invalid entries)");
490 return -EIO;
492 if (MSDOS_I(dir)->i_start == 0 ||
493 MSDOS_I(dir)->i_start == MSDOS_SB(sb)->root_cluster) {
494 /* Directory should point valid cluster. */
495 fat_fs_error(sb, "corrupted directory (invalid i_start)");
496 return -EIO;
498 return 0;
501 /* doesn't deal with root inode */
502 int fat_fill_inode(struct inode *inode, struct msdos_dir_entry *de)
504 struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
505 int error;
507 MSDOS_I(inode)->i_pos = 0;
508 inode->i_uid = sbi->options.fs_uid;
509 inode->i_gid = sbi->options.fs_gid;
510 inode->i_version++;
511 inode->i_generation = get_seconds();
513 if ((de->attr & ATTR_DIR) && !IS_FREE(de->name)) {
514 inode->i_generation &= ~1;
515 inode->i_mode = fat_make_mode(sbi, de->attr, S_IRWXUGO);
516 inode->i_op = sbi->dir_ops;
517 inode->i_fop = &fat_dir_operations;
519 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
520 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
521 error = fat_calc_dir_size(inode);
522 if (error < 0)
523 return error;
524 MSDOS_I(inode)->mmu_private = inode->i_size;
526 set_nlink(inode, fat_subdirs(inode));
528 error = fat_validate_dir(inode);
529 if (error < 0)
530 return error;
531 } else { /* not a directory */
532 inode->i_generation |= 1;
533 inode->i_mode = fat_make_mode(sbi, de->attr,
534 ((sbi->options.showexec && !is_exec(de->name + 8))
535 ? S_IRUGO|S_IWUGO : S_IRWXUGO));
536 MSDOS_I(inode)->i_start = fat_get_start(sbi, de);
538 MSDOS_I(inode)->i_logstart = MSDOS_I(inode)->i_start;
539 inode->i_size = le32_to_cpu(de->size);
540 inode->i_op = &fat_file_inode_operations;
541 inode->i_fop = &fat_file_operations;
542 inode->i_mapping->a_ops = &fat_aops;
543 MSDOS_I(inode)->mmu_private = inode->i_size;
545 if (de->attr & ATTR_SYS) {
546 if (sbi->options.sys_immutable)
547 inode->i_flags |= S_IMMUTABLE;
549 fat_save_attrs(inode, de->attr);
551 inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
552 & ~((loff_t)sbi->cluster_size - 1)) >> 9;
554 fat_time_fat2unix(sbi, &inode->i_mtime, de->time, de->date, 0);
555 if (sbi->options.isvfat) {
556 fat_time_fat2unix(sbi, &inode->i_ctime, de->ctime,
557 de->cdate, de->ctime_cs);
558 fat_time_fat2unix(sbi, &inode->i_atime, 0, de->adate, 0);
559 } else
560 inode->i_ctime = inode->i_atime = inode->i_mtime;
562 return 0;
565 static inline void fat_lock_build_inode(struct msdos_sb_info *sbi)
567 if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
568 mutex_lock(&sbi->nfs_build_inode_lock);
571 static inline void fat_unlock_build_inode(struct msdos_sb_info *sbi)
573 if (sbi->options.nfs == FAT_NFS_NOSTALE_RO)
574 mutex_unlock(&sbi->nfs_build_inode_lock);
577 struct inode *fat_build_inode(struct super_block *sb,
578 struct msdos_dir_entry *de, loff_t i_pos)
580 struct inode *inode;
581 int err;
583 fat_lock_build_inode(MSDOS_SB(sb));
584 inode = fat_iget(sb, i_pos);
585 if (inode)
586 goto out;
587 inode = new_inode(sb);
588 if (!inode) {
589 inode = ERR_PTR(-ENOMEM);
590 goto out;
592 inode->i_ino = iunique(sb, MSDOS_ROOT_INO);
593 inode->i_version = 1;
594 err = fat_fill_inode(inode, de);
595 if (err) {
596 iput(inode);
597 inode = ERR_PTR(err);
598 goto out;
600 fat_attach(inode, i_pos);
601 insert_inode_hash(inode);
602 out:
603 fat_unlock_build_inode(MSDOS_SB(sb));
604 return inode;
607 EXPORT_SYMBOL_GPL(fat_build_inode);
609 static int __fat_write_inode(struct inode *inode, int wait);
611 static void fat_free_eofblocks(struct inode *inode)
613 /* Release unwritten fallocated blocks on inode eviction. */
614 if ((inode->i_blocks << 9) >
615 round_up(MSDOS_I(inode)->mmu_private,
616 MSDOS_SB(inode->i_sb)->cluster_size)) {
617 int err;
619 fat_truncate_blocks(inode, MSDOS_I(inode)->mmu_private);
620 /* Fallocate results in updating the i_start/iogstart
621 * for the zero byte file. So, make it return to
622 * original state during evict and commit it to avoid
623 * any corruption on the next access to the cluster
624 * chain for the file.
626 err = __fat_write_inode(inode, inode_needs_sync(inode));
627 if (err) {
628 fat_msg(inode->i_sb, KERN_WARNING, "Failed to "
629 "update on disk inode for unused "
630 "fallocated blocks, inode could be "
631 "corrupted. Please run fsck");
637 static void fat_evict_inode(struct inode *inode)
639 truncate_inode_pages_final(&inode->i_data);
640 if (!inode->i_nlink) {
641 inode->i_size = 0;
642 fat_truncate_blocks(inode, 0);
643 } else
644 fat_free_eofblocks(inode);
646 invalidate_inode_buffers(inode);
647 clear_inode(inode);
648 fat_cache_inval_inode(inode);
649 fat_detach(inode);
652 static void fat_set_state(struct super_block *sb,
653 unsigned int set, unsigned int force)
655 struct buffer_head *bh;
656 struct fat_boot_sector *b;
657 struct msdos_sb_info *sbi = MSDOS_SB(sb);
659 /* do not change any thing if mounted read only */
660 if (sb_rdonly(sb) && !force)
661 return;
663 /* do not change state if fs was dirty */
664 if (sbi->dirty) {
665 /* warn only on set (mount). */
666 if (set)
667 fat_msg(sb, KERN_WARNING, "Volume was not properly "
668 "unmounted. Some data may be corrupt. "
669 "Please run fsck.");
670 return;
673 bh = sb_bread(sb, 0);
674 if (bh == NULL) {
675 fat_msg(sb, KERN_ERR, "unable to read boot sector "
676 "to mark fs as dirty");
677 return;
680 b = (struct fat_boot_sector *) bh->b_data;
682 if (sbi->fat_bits == 32) {
683 if (set)
684 b->fat32.state |= FAT_STATE_DIRTY;
685 else
686 b->fat32.state &= ~FAT_STATE_DIRTY;
687 } else /* fat 16 and 12 */ {
688 if (set)
689 b->fat16.state |= FAT_STATE_DIRTY;
690 else
691 b->fat16.state &= ~FAT_STATE_DIRTY;
694 mark_buffer_dirty(bh);
695 sync_dirty_buffer(bh);
696 brelse(bh);
699 static void fat_reset_iocharset(struct fat_mount_options *opts)
701 if (opts->iocharset != fat_default_iocharset) {
702 /* Note: opts->iocharset can be NULL here */
703 kfree(opts->iocharset);
704 opts->iocharset = fat_default_iocharset;
708 static void delayed_free(struct rcu_head *p)
710 struct msdos_sb_info *sbi = container_of(p, struct msdos_sb_info, rcu);
711 unload_nls(sbi->nls_disk);
712 unload_nls(sbi->nls_io);
713 fat_reset_iocharset(&sbi->options);
714 kfree(sbi);
717 static void fat_put_super(struct super_block *sb)
719 struct msdos_sb_info *sbi = MSDOS_SB(sb);
721 fat_set_state(sb, 0, 0);
723 iput(sbi->fsinfo_inode);
724 iput(sbi->fat_inode);
726 call_rcu(&sbi->rcu, delayed_free);
729 static struct kmem_cache *fat_inode_cachep;
731 static struct inode *fat_alloc_inode(struct super_block *sb)
733 struct msdos_inode_info *ei;
734 ei = kmem_cache_alloc(fat_inode_cachep, GFP_NOFS);
735 if (!ei)
736 return NULL;
738 init_rwsem(&ei->truncate_lock);
739 /* Zeroing to allow iput() even if partial initialized inode. */
740 ei->mmu_private = 0;
741 ei->i_start = 0;
742 ei->i_logstart = 0;
743 ei->i_attrs = 0;
744 ei->i_pos = 0;
746 return &ei->vfs_inode;
749 static void fat_i_callback(struct rcu_head *head)
751 struct inode *inode = container_of(head, struct inode, i_rcu);
752 kmem_cache_free(fat_inode_cachep, MSDOS_I(inode));
755 static void fat_destroy_inode(struct inode *inode)
757 call_rcu(&inode->i_rcu, fat_i_callback);
760 static void init_once(void *foo)
762 struct msdos_inode_info *ei = (struct msdos_inode_info *)foo;
764 spin_lock_init(&ei->cache_lru_lock);
765 ei->nr_caches = 0;
766 ei->cache_valid_id = FAT_CACHE_VALID + 1;
767 INIT_LIST_HEAD(&ei->cache_lru);
768 INIT_HLIST_NODE(&ei->i_fat_hash);
769 INIT_HLIST_NODE(&ei->i_dir_hash);
770 inode_init_once(&ei->vfs_inode);
773 static int __init fat_init_inodecache(void)
775 fat_inode_cachep = kmem_cache_create("fat_inode_cache",
776 sizeof(struct msdos_inode_info),
777 0, (SLAB_RECLAIM_ACCOUNT|
778 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
779 init_once);
780 if (fat_inode_cachep == NULL)
781 return -ENOMEM;
782 return 0;
785 static void __exit fat_destroy_inodecache(void)
788 * Make sure all delayed rcu free inodes are flushed before we
789 * destroy cache.
791 rcu_barrier();
792 kmem_cache_destroy(fat_inode_cachep);
795 static int fat_remount(struct super_block *sb, int *flags, char *data)
797 bool new_rdonly;
798 struct msdos_sb_info *sbi = MSDOS_SB(sb);
799 *flags |= MS_NODIRATIME | (sbi->options.isvfat ? 0 : MS_NOATIME);
801 sync_filesystem(sb);
803 /* make sure we update state on remount. */
804 new_rdonly = *flags & MS_RDONLY;
805 if (new_rdonly != sb_rdonly(sb)) {
806 if (new_rdonly)
807 fat_set_state(sb, 0, 0);
808 else
809 fat_set_state(sb, 1, 1);
811 return 0;
814 static int fat_statfs(struct dentry *dentry, struct kstatfs *buf)
816 struct super_block *sb = dentry->d_sb;
817 struct msdos_sb_info *sbi = MSDOS_SB(sb);
818 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
820 /* If the count of free cluster is still unknown, counts it here. */
821 if (sbi->free_clusters == -1 || !sbi->free_clus_valid) {
822 int err = fat_count_free_clusters(dentry->d_sb);
823 if (err)
824 return err;
827 buf->f_type = dentry->d_sb->s_magic;
828 buf->f_bsize = sbi->cluster_size;
829 buf->f_blocks = sbi->max_cluster - FAT_START_ENT;
830 buf->f_bfree = sbi->free_clusters;
831 buf->f_bavail = sbi->free_clusters;
832 buf->f_fsid.val[0] = (u32)id;
833 buf->f_fsid.val[1] = (u32)(id >> 32);
834 buf->f_namelen =
835 (sbi->options.isvfat ? FAT_LFN_LEN : 12) * NLS_MAX_CHARSET_SIZE;
837 return 0;
840 static int __fat_write_inode(struct inode *inode, int wait)
842 struct super_block *sb = inode->i_sb;
843 struct msdos_sb_info *sbi = MSDOS_SB(sb);
844 struct buffer_head *bh;
845 struct msdos_dir_entry *raw_entry;
846 loff_t i_pos;
847 sector_t blocknr;
848 int err, offset;
850 if (inode->i_ino == MSDOS_ROOT_INO)
851 return 0;
853 retry:
854 i_pos = fat_i_pos_read(sbi, inode);
855 if (!i_pos)
856 return 0;
858 fat_get_blknr_offset(sbi, i_pos, &blocknr, &offset);
859 bh = sb_bread(sb, blocknr);
860 if (!bh) {
861 fat_msg(sb, KERN_ERR, "unable to read inode block "
862 "for updating (i_pos %lld)", i_pos);
863 return -EIO;
865 spin_lock(&sbi->inode_hash_lock);
866 if (i_pos != MSDOS_I(inode)->i_pos) {
867 spin_unlock(&sbi->inode_hash_lock);
868 brelse(bh);
869 goto retry;
872 raw_entry = &((struct msdos_dir_entry *) (bh->b_data))[offset];
873 if (S_ISDIR(inode->i_mode))
874 raw_entry->size = 0;
875 else
876 raw_entry->size = cpu_to_le32(inode->i_size);
877 raw_entry->attr = fat_make_attrs(inode);
878 fat_set_start(raw_entry, MSDOS_I(inode)->i_logstart);
879 fat_time_unix2fat(sbi, &inode->i_mtime, &raw_entry->time,
880 &raw_entry->date, NULL);
881 if (sbi->options.isvfat) {
882 __le16 atime;
883 fat_time_unix2fat(sbi, &inode->i_ctime, &raw_entry->ctime,
884 &raw_entry->cdate, &raw_entry->ctime_cs);
885 fat_time_unix2fat(sbi, &inode->i_atime, &atime,
886 &raw_entry->adate, NULL);
888 spin_unlock(&sbi->inode_hash_lock);
889 mark_buffer_dirty(bh);
890 err = 0;
891 if (wait)
892 err = sync_dirty_buffer(bh);
893 brelse(bh);
894 return err;
897 static int fat_write_inode(struct inode *inode, struct writeback_control *wbc)
899 int err;
901 if (inode->i_ino == MSDOS_FSINFO_INO) {
902 struct super_block *sb = inode->i_sb;
904 mutex_lock(&MSDOS_SB(sb)->s_lock);
905 err = fat_clusters_flush(sb);
906 mutex_unlock(&MSDOS_SB(sb)->s_lock);
907 } else
908 err = __fat_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
910 return err;
913 int fat_sync_inode(struct inode *inode)
915 return __fat_write_inode(inode, 1);
918 EXPORT_SYMBOL_GPL(fat_sync_inode);
920 static int fat_show_options(struct seq_file *m, struct dentry *root);
921 static const struct super_operations fat_sops = {
922 .alloc_inode = fat_alloc_inode,
923 .destroy_inode = fat_destroy_inode,
924 .write_inode = fat_write_inode,
925 .evict_inode = fat_evict_inode,
926 .put_super = fat_put_super,
927 .statfs = fat_statfs,
928 .remount_fs = fat_remount,
930 .show_options = fat_show_options,
933 static int fat_show_options(struct seq_file *m, struct dentry *root)
935 struct msdos_sb_info *sbi = MSDOS_SB(root->d_sb);
936 struct fat_mount_options *opts = &sbi->options;
937 int isvfat = opts->isvfat;
939 if (!uid_eq(opts->fs_uid, GLOBAL_ROOT_UID))
940 seq_printf(m, ",uid=%u",
941 from_kuid_munged(&init_user_ns, opts->fs_uid));
942 if (!gid_eq(opts->fs_gid, GLOBAL_ROOT_GID))
943 seq_printf(m, ",gid=%u",
944 from_kgid_munged(&init_user_ns, opts->fs_gid));
945 seq_printf(m, ",fmask=%04o", opts->fs_fmask);
946 seq_printf(m, ",dmask=%04o", opts->fs_dmask);
947 if (opts->allow_utime)
948 seq_printf(m, ",allow_utime=%04o", opts->allow_utime);
949 if (sbi->nls_disk)
950 /* strip "cp" prefix from displayed option */
951 seq_printf(m, ",codepage=%s", &sbi->nls_disk->charset[2]);
952 if (isvfat) {
953 if (sbi->nls_io)
954 seq_printf(m, ",iocharset=%s", sbi->nls_io->charset);
956 switch (opts->shortname) {
957 case VFAT_SFN_DISPLAY_WIN95 | VFAT_SFN_CREATE_WIN95:
958 seq_puts(m, ",shortname=win95");
959 break;
960 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WINNT:
961 seq_puts(m, ",shortname=winnt");
962 break;
963 case VFAT_SFN_DISPLAY_WINNT | VFAT_SFN_CREATE_WIN95:
964 seq_puts(m, ",shortname=mixed");
965 break;
966 case VFAT_SFN_DISPLAY_LOWER | VFAT_SFN_CREATE_WIN95:
967 seq_puts(m, ",shortname=lower");
968 break;
969 default:
970 seq_puts(m, ",shortname=unknown");
971 break;
974 if (opts->name_check != 'n')
975 seq_printf(m, ",check=%c", opts->name_check);
976 if (opts->usefree)
977 seq_puts(m, ",usefree");
978 if (opts->quiet)
979 seq_puts(m, ",quiet");
980 if (opts->showexec)
981 seq_puts(m, ",showexec");
982 if (opts->sys_immutable)
983 seq_puts(m, ",sys_immutable");
984 if (!isvfat) {
985 if (opts->dotsOK)
986 seq_puts(m, ",dotsOK=yes");
987 if (opts->nocase)
988 seq_puts(m, ",nocase");
989 } else {
990 if (opts->utf8)
991 seq_puts(m, ",utf8");
992 if (opts->unicode_xlate)
993 seq_puts(m, ",uni_xlate");
994 if (!opts->numtail)
995 seq_puts(m, ",nonumtail");
996 if (opts->rodir)
997 seq_puts(m, ",rodir");
999 if (opts->flush)
1000 seq_puts(m, ",flush");
1001 if (opts->tz_set) {
1002 if (opts->time_offset)
1003 seq_printf(m, ",time_offset=%d", opts->time_offset);
1004 else
1005 seq_puts(m, ",tz=UTC");
1007 if (opts->errors == FAT_ERRORS_CONT)
1008 seq_puts(m, ",errors=continue");
1009 else if (opts->errors == FAT_ERRORS_PANIC)
1010 seq_puts(m, ",errors=panic");
1011 else
1012 seq_puts(m, ",errors=remount-ro");
1013 if (opts->nfs == FAT_NFS_NOSTALE_RO)
1014 seq_puts(m, ",nfs=nostale_ro");
1015 else if (opts->nfs)
1016 seq_puts(m, ",nfs=stale_rw");
1017 if (opts->discard)
1018 seq_puts(m, ",discard");
1019 if (opts->dos1xfloppy)
1020 seq_puts(m, ",dos1xfloppy");
1022 return 0;
1025 enum {
1026 Opt_check_n, Opt_check_r, Opt_check_s, Opt_uid, Opt_gid,
1027 Opt_umask, Opt_dmask, Opt_fmask, Opt_allow_utime, Opt_codepage,
1028 Opt_usefree, Opt_nocase, Opt_quiet, Opt_showexec, Opt_debug,
1029 Opt_immutable, Opt_dots, Opt_nodots,
1030 Opt_charset, Opt_shortname_lower, Opt_shortname_win95,
1031 Opt_shortname_winnt, Opt_shortname_mixed, Opt_utf8_no, Opt_utf8_yes,
1032 Opt_uni_xl_no, Opt_uni_xl_yes, Opt_nonumtail_no, Opt_nonumtail_yes,
1033 Opt_obsolete, Opt_flush, Opt_tz_utc, Opt_rodir, Opt_err_cont,
1034 Opt_err_panic, Opt_err_ro, Opt_discard, Opt_nfs, Opt_time_offset,
1035 Opt_nfs_stale_rw, Opt_nfs_nostale_ro, Opt_err, Opt_dos1xfloppy,
1038 static const match_table_t fat_tokens = {
1039 {Opt_check_r, "check=relaxed"},
1040 {Opt_check_s, "check=strict"},
1041 {Opt_check_n, "check=normal"},
1042 {Opt_check_r, "check=r"},
1043 {Opt_check_s, "check=s"},
1044 {Opt_check_n, "check=n"},
1045 {Opt_uid, "uid=%u"},
1046 {Opt_gid, "gid=%u"},
1047 {Opt_umask, "umask=%o"},
1048 {Opt_dmask, "dmask=%o"},
1049 {Opt_fmask, "fmask=%o"},
1050 {Opt_allow_utime, "allow_utime=%o"},
1051 {Opt_codepage, "codepage=%u"},
1052 {Opt_usefree, "usefree"},
1053 {Opt_nocase, "nocase"},
1054 {Opt_quiet, "quiet"},
1055 {Opt_showexec, "showexec"},
1056 {Opt_debug, "debug"},
1057 {Opt_immutable, "sys_immutable"},
1058 {Opt_flush, "flush"},
1059 {Opt_tz_utc, "tz=UTC"},
1060 {Opt_time_offset, "time_offset=%d"},
1061 {Opt_err_cont, "errors=continue"},
1062 {Opt_err_panic, "errors=panic"},
1063 {Opt_err_ro, "errors=remount-ro"},
1064 {Opt_discard, "discard"},
1065 {Opt_nfs_stale_rw, "nfs"},
1066 {Opt_nfs_stale_rw, "nfs=stale_rw"},
1067 {Opt_nfs_nostale_ro, "nfs=nostale_ro"},
1068 {Opt_dos1xfloppy, "dos1xfloppy"},
1069 {Opt_obsolete, "conv=binary"},
1070 {Opt_obsolete, "conv=text"},
1071 {Opt_obsolete, "conv=auto"},
1072 {Opt_obsolete, "conv=b"},
1073 {Opt_obsolete, "conv=t"},
1074 {Opt_obsolete, "conv=a"},
1075 {Opt_obsolete, "fat=%u"},
1076 {Opt_obsolete, "blocksize=%u"},
1077 {Opt_obsolete, "cvf_format=%20s"},
1078 {Opt_obsolete, "cvf_options=%100s"},
1079 {Opt_obsolete, "posix"},
1080 {Opt_err, NULL},
1082 static const match_table_t msdos_tokens = {
1083 {Opt_nodots, "nodots"},
1084 {Opt_nodots, "dotsOK=no"},
1085 {Opt_dots, "dots"},
1086 {Opt_dots, "dotsOK=yes"},
1087 {Opt_err, NULL}
1089 static const match_table_t vfat_tokens = {
1090 {Opt_charset, "iocharset=%s"},
1091 {Opt_shortname_lower, "shortname=lower"},
1092 {Opt_shortname_win95, "shortname=win95"},
1093 {Opt_shortname_winnt, "shortname=winnt"},
1094 {Opt_shortname_mixed, "shortname=mixed"},
1095 {Opt_utf8_no, "utf8=0"}, /* 0 or no or false */
1096 {Opt_utf8_no, "utf8=no"},
1097 {Opt_utf8_no, "utf8=false"},
1098 {Opt_utf8_yes, "utf8=1"}, /* empty or 1 or yes or true */
1099 {Opt_utf8_yes, "utf8=yes"},
1100 {Opt_utf8_yes, "utf8=true"},
1101 {Opt_utf8_yes, "utf8"},
1102 {Opt_uni_xl_no, "uni_xlate=0"}, /* 0 or no or false */
1103 {Opt_uni_xl_no, "uni_xlate=no"},
1104 {Opt_uni_xl_no, "uni_xlate=false"},
1105 {Opt_uni_xl_yes, "uni_xlate=1"}, /* empty or 1 or yes or true */
1106 {Opt_uni_xl_yes, "uni_xlate=yes"},
1107 {Opt_uni_xl_yes, "uni_xlate=true"},
1108 {Opt_uni_xl_yes, "uni_xlate"},
1109 {Opt_nonumtail_no, "nonumtail=0"}, /* 0 or no or false */
1110 {Opt_nonumtail_no, "nonumtail=no"},
1111 {Opt_nonumtail_no, "nonumtail=false"},
1112 {Opt_nonumtail_yes, "nonumtail=1"}, /* empty or 1 or yes or true */
1113 {Opt_nonumtail_yes, "nonumtail=yes"},
1114 {Opt_nonumtail_yes, "nonumtail=true"},
1115 {Opt_nonumtail_yes, "nonumtail"},
1116 {Opt_rodir, "rodir"},
1117 {Opt_err, NULL}
1120 static int parse_options(struct super_block *sb, char *options, int is_vfat,
1121 int silent, int *debug, struct fat_mount_options *opts)
1123 char *p;
1124 substring_t args[MAX_OPT_ARGS];
1125 int option;
1126 char *iocharset;
1128 opts->isvfat = is_vfat;
1130 opts->fs_uid = current_uid();
1131 opts->fs_gid = current_gid();
1132 opts->fs_fmask = opts->fs_dmask = current_umask();
1133 opts->allow_utime = -1;
1134 opts->codepage = fat_default_codepage;
1135 fat_reset_iocharset(opts);
1136 if (is_vfat) {
1137 opts->shortname = VFAT_SFN_DISPLAY_WINNT|VFAT_SFN_CREATE_WIN95;
1138 opts->rodir = 0;
1139 } else {
1140 opts->shortname = 0;
1141 opts->rodir = 1;
1143 opts->name_check = 'n';
1144 opts->quiet = opts->showexec = opts->sys_immutable = opts->dotsOK = 0;
1145 opts->unicode_xlate = 0;
1146 opts->numtail = 1;
1147 opts->usefree = opts->nocase = 0;
1148 opts->tz_set = 0;
1149 opts->nfs = 0;
1150 opts->errors = FAT_ERRORS_RO;
1151 *debug = 0;
1153 opts->utf8 = IS_ENABLED(CONFIG_FAT_DEFAULT_UTF8) && is_vfat;
1155 if (!options)
1156 goto out;
1158 while ((p = strsep(&options, ",")) != NULL) {
1159 int token;
1160 if (!*p)
1161 continue;
1163 token = match_token(p, fat_tokens, args);
1164 if (token == Opt_err) {
1165 if (is_vfat)
1166 token = match_token(p, vfat_tokens, args);
1167 else
1168 token = match_token(p, msdos_tokens, args);
1170 switch (token) {
1171 case Opt_check_s:
1172 opts->name_check = 's';
1173 break;
1174 case Opt_check_r:
1175 opts->name_check = 'r';
1176 break;
1177 case Opt_check_n:
1178 opts->name_check = 'n';
1179 break;
1180 case Opt_usefree:
1181 opts->usefree = 1;
1182 break;
1183 case Opt_nocase:
1184 if (!is_vfat)
1185 opts->nocase = 1;
1186 else {
1187 /* for backward compatibility */
1188 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1189 | VFAT_SFN_CREATE_WIN95;
1191 break;
1192 case Opt_quiet:
1193 opts->quiet = 1;
1194 break;
1195 case Opt_showexec:
1196 opts->showexec = 1;
1197 break;
1198 case Opt_debug:
1199 *debug = 1;
1200 break;
1201 case Opt_immutable:
1202 opts->sys_immutable = 1;
1203 break;
1204 case Opt_uid:
1205 if (match_int(&args[0], &option))
1206 return -EINVAL;
1207 opts->fs_uid = make_kuid(current_user_ns(), option);
1208 if (!uid_valid(opts->fs_uid))
1209 return -EINVAL;
1210 break;
1211 case Opt_gid:
1212 if (match_int(&args[0], &option))
1213 return -EINVAL;
1214 opts->fs_gid = make_kgid(current_user_ns(), option);
1215 if (!gid_valid(opts->fs_gid))
1216 return -EINVAL;
1217 break;
1218 case Opt_umask:
1219 if (match_octal(&args[0], &option))
1220 return -EINVAL;
1221 opts->fs_fmask = opts->fs_dmask = option;
1222 break;
1223 case Opt_dmask:
1224 if (match_octal(&args[0], &option))
1225 return -EINVAL;
1226 opts->fs_dmask = option;
1227 break;
1228 case Opt_fmask:
1229 if (match_octal(&args[0], &option))
1230 return -EINVAL;
1231 opts->fs_fmask = option;
1232 break;
1233 case Opt_allow_utime:
1234 if (match_octal(&args[0], &option))
1235 return -EINVAL;
1236 opts->allow_utime = option & (S_IWGRP | S_IWOTH);
1237 break;
1238 case Opt_codepage:
1239 if (match_int(&args[0], &option))
1240 return -EINVAL;
1241 opts->codepage = option;
1242 break;
1243 case Opt_flush:
1244 opts->flush = 1;
1245 break;
1246 case Opt_time_offset:
1247 if (match_int(&args[0], &option))
1248 return -EINVAL;
1250 * GMT+-12 zones may have DST corrections so at least
1251 * 13 hours difference is needed. Make the limit 24
1252 * just in case someone invents something unusual.
1254 if (option < -24 * 60 || option > 24 * 60)
1255 return -EINVAL;
1256 opts->tz_set = 1;
1257 opts->time_offset = option;
1258 break;
1259 case Opt_tz_utc:
1260 opts->tz_set = 1;
1261 opts->time_offset = 0;
1262 break;
1263 case Opt_err_cont:
1264 opts->errors = FAT_ERRORS_CONT;
1265 break;
1266 case Opt_err_panic:
1267 opts->errors = FAT_ERRORS_PANIC;
1268 break;
1269 case Opt_err_ro:
1270 opts->errors = FAT_ERRORS_RO;
1271 break;
1272 case Opt_nfs_stale_rw:
1273 opts->nfs = FAT_NFS_STALE_RW;
1274 break;
1275 case Opt_nfs_nostale_ro:
1276 opts->nfs = FAT_NFS_NOSTALE_RO;
1277 break;
1278 case Opt_dos1xfloppy:
1279 opts->dos1xfloppy = 1;
1280 break;
1282 /* msdos specific */
1283 case Opt_dots:
1284 opts->dotsOK = 1;
1285 break;
1286 case Opt_nodots:
1287 opts->dotsOK = 0;
1288 break;
1290 /* vfat specific */
1291 case Opt_charset:
1292 fat_reset_iocharset(opts);
1293 iocharset = match_strdup(&args[0]);
1294 if (!iocharset)
1295 return -ENOMEM;
1296 opts->iocharset = iocharset;
1297 break;
1298 case Opt_shortname_lower:
1299 opts->shortname = VFAT_SFN_DISPLAY_LOWER
1300 | VFAT_SFN_CREATE_WIN95;
1301 break;
1302 case Opt_shortname_win95:
1303 opts->shortname = VFAT_SFN_DISPLAY_WIN95
1304 | VFAT_SFN_CREATE_WIN95;
1305 break;
1306 case Opt_shortname_winnt:
1307 opts->shortname = VFAT_SFN_DISPLAY_WINNT
1308 | VFAT_SFN_CREATE_WINNT;
1309 break;
1310 case Opt_shortname_mixed:
1311 opts->shortname = VFAT_SFN_DISPLAY_WINNT
1312 | VFAT_SFN_CREATE_WIN95;
1313 break;
1314 case Opt_utf8_no: /* 0 or no or false */
1315 opts->utf8 = 0;
1316 break;
1317 case Opt_utf8_yes: /* empty or 1 or yes or true */
1318 opts->utf8 = 1;
1319 break;
1320 case Opt_uni_xl_no: /* 0 or no or false */
1321 opts->unicode_xlate = 0;
1322 break;
1323 case Opt_uni_xl_yes: /* empty or 1 or yes or true */
1324 opts->unicode_xlate = 1;
1325 break;
1326 case Opt_nonumtail_no: /* 0 or no or false */
1327 opts->numtail = 1; /* negated option */
1328 break;
1329 case Opt_nonumtail_yes: /* empty or 1 or yes or true */
1330 opts->numtail = 0; /* negated option */
1331 break;
1332 case Opt_rodir:
1333 opts->rodir = 1;
1334 break;
1335 case Opt_discard:
1336 opts->discard = 1;
1337 break;
1339 /* obsolete mount options */
1340 case Opt_obsolete:
1341 fat_msg(sb, KERN_INFO, "\"%s\" option is obsolete, "
1342 "not supported now", p);
1343 break;
1344 /* unknown option */
1345 default:
1346 if (!silent) {
1347 fat_msg(sb, KERN_ERR,
1348 "Unrecognized mount option \"%s\" "
1349 "or missing value", p);
1351 return -EINVAL;
1355 out:
1356 /* UTF-8 doesn't provide FAT semantics */
1357 if (!strcmp(opts->iocharset, "utf8")) {
1358 fat_msg(sb, KERN_WARNING, "utf8 is not a recommended IO charset"
1359 " for FAT filesystems, filesystem will be "
1360 "case sensitive!");
1363 /* If user doesn't specify allow_utime, it's initialized from dmask. */
1364 if (opts->allow_utime == (unsigned short)-1)
1365 opts->allow_utime = ~opts->fs_dmask & (S_IWGRP | S_IWOTH);
1366 if (opts->unicode_xlate)
1367 opts->utf8 = 0;
1368 if (opts->nfs == FAT_NFS_NOSTALE_RO) {
1369 sb->s_flags |= MS_RDONLY;
1370 sb->s_export_op = &fat_export_ops_nostale;
1373 return 0;
1376 static int fat_read_root(struct inode *inode)
1378 struct msdos_sb_info *sbi = MSDOS_SB(inode->i_sb);
1379 int error;
1381 MSDOS_I(inode)->i_pos = MSDOS_ROOT_INO;
1382 inode->i_uid = sbi->options.fs_uid;
1383 inode->i_gid = sbi->options.fs_gid;
1384 inode->i_version++;
1385 inode->i_generation = 0;
1386 inode->i_mode = fat_make_mode(sbi, ATTR_DIR, S_IRWXUGO);
1387 inode->i_op = sbi->dir_ops;
1388 inode->i_fop = &fat_dir_operations;
1389 if (sbi->fat_bits == 32) {
1390 MSDOS_I(inode)->i_start = sbi->root_cluster;
1391 error = fat_calc_dir_size(inode);
1392 if (error < 0)
1393 return error;
1394 } else {
1395 MSDOS_I(inode)->i_start = 0;
1396 inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
1398 inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
1399 & ~((loff_t)sbi->cluster_size - 1)) >> 9;
1400 MSDOS_I(inode)->i_logstart = 0;
1401 MSDOS_I(inode)->mmu_private = inode->i_size;
1403 fat_save_attrs(inode, ATTR_DIR);
1404 inode->i_mtime.tv_sec = inode->i_atime.tv_sec = inode->i_ctime.tv_sec = 0;
1405 inode->i_mtime.tv_nsec = inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = 0;
1406 set_nlink(inode, fat_subdirs(inode)+2);
1408 return 0;
1411 static unsigned long calc_fat_clusters(struct super_block *sb)
1413 struct msdos_sb_info *sbi = MSDOS_SB(sb);
1415 /* Divide first to avoid overflow */
1416 if (sbi->fat_bits != 12) {
1417 unsigned long ent_per_sec = sb->s_blocksize * 8 / sbi->fat_bits;
1418 return ent_per_sec * sbi->fat_length;
1421 return sbi->fat_length * sb->s_blocksize * 8 / sbi->fat_bits;
1424 static bool fat_bpb_is_zero(struct fat_boot_sector *b)
1426 if (get_unaligned_le16(&b->sector_size))
1427 return false;
1428 if (b->sec_per_clus)
1429 return false;
1430 if (b->reserved)
1431 return false;
1432 if (b->fats)
1433 return false;
1434 if (get_unaligned_le16(&b->dir_entries))
1435 return false;
1436 if (get_unaligned_le16(&b->sectors))
1437 return false;
1438 if (b->media)
1439 return false;
1440 if (b->fat_length)
1441 return false;
1442 if (b->secs_track)
1443 return false;
1444 if (b->heads)
1445 return false;
1446 return true;
1449 static int fat_read_bpb(struct super_block *sb, struct fat_boot_sector *b,
1450 int silent, struct fat_bios_param_block *bpb)
1452 int error = -EINVAL;
1454 /* Read in BPB ... */
1455 memset(bpb, 0, sizeof(*bpb));
1456 bpb->fat_sector_size = get_unaligned_le16(&b->sector_size);
1457 bpb->fat_sec_per_clus = b->sec_per_clus;
1458 bpb->fat_reserved = le16_to_cpu(b->reserved);
1459 bpb->fat_fats = b->fats;
1460 bpb->fat_dir_entries = get_unaligned_le16(&b->dir_entries);
1461 bpb->fat_sectors = get_unaligned_le16(&b->sectors);
1462 bpb->fat_fat_length = le16_to_cpu(b->fat_length);
1463 bpb->fat_total_sect = le32_to_cpu(b->total_sect);
1465 bpb->fat16_state = b->fat16.state;
1466 bpb->fat16_vol_id = get_unaligned_le32(b->fat16.vol_id);
1468 bpb->fat32_length = le32_to_cpu(b->fat32.length);
1469 bpb->fat32_root_cluster = le32_to_cpu(b->fat32.root_cluster);
1470 bpb->fat32_info_sector = le16_to_cpu(b->fat32.info_sector);
1471 bpb->fat32_state = b->fat32.state;
1472 bpb->fat32_vol_id = get_unaligned_le32(b->fat32.vol_id);
1474 /* Validate this looks like a FAT filesystem BPB */
1475 if (!bpb->fat_reserved) {
1476 if (!silent)
1477 fat_msg(sb, KERN_ERR,
1478 "bogus number of reserved sectors");
1479 goto out;
1481 if (!bpb->fat_fats) {
1482 if (!silent)
1483 fat_msg(sb, KERN_ERR, "bogus number of FAT structure");
1484 goto out;
1488 * Earlier we checked here that b->secs_track and b->head are nonzero,
1489 * but it turns out valid FAT filesystems can have zero there.
1492 if (!fat_valid_media(b->media)) {
1493 if (!silent)
1494 fat_msg(sb, KERN_ERR, "invalid media value (0x%02x)",
1495 (unsigned)b->media);
1496 goto out;
1499 if (!is_power_of_2(bpb->fat_sector_size)
1500 || (bpb->fat_sector_size < 512)
1501 || (bpb->fat_sector_size > 4096)) {
1502 if (!silent)
1503 fat_msg(sb, KERN_ERR, "bogus logical sector size %u",
1504 (unsigned)bpb->fat_sector_size);
1505 goto out;
1508 if (!is_power_of_2(bpb->fat_sec_per_clus)) {
1509 if (!silent)
1510 fat_msg(sb, KERN_ERR, "bogus sectors per cluster %u",
1511 (unsigned)bpb->fat_sec_per_clus);
1512 goto out;
1515 if (bpb->fat_fat_length == 0 && bpb->fat32_length == 0) {
1516 if (!silent)
1517 fat_msg(sb, KERN_ERR, "bogus number of FAT sectors");
1518 goto out;
1521 error = 0;
1523 out:
1524 return error;
1527 static int fat_read_static_bpb(struct super_block *sb,
1528 struct fat_boot_sector *b, int silent,
1529 struct fat_bios_param_block *bpb)
1531 static const char *notdos1x = "This doesn't look like a DOS 1.x volume";
1533 struct fat_floppy_defaults *fdefaults = NULL;
1534 int error = -EINVAL;
1535 sector_t bd_sects;
1536 unsigned i;
1538 bd_sects = i_size_read(sb->s_bdev->bd_inode) / SECTOR_SIZE;
1540 /* 16-bit DOS 1.x reliably wrote bootstrap short-jmp code */
1541 if (b->ignored[0] != 0xeb || b->ignored[2] != 0x90) {
1542 if (!silent)
1543 fat_msg(sb, KERN_ERR,
1544 "%s; no bootstrapping code", notdos1x);
1545 goto out;
1549 * If any value in this region is non-zero, it isn't archaic
1550 * DOS.
1552 if (!fat_bpb_is_zero(b)) {
1553 if (!silent)
1554 fat_msg(sb, KERN_ERR,
1555 "%s; DOS 2.x BPB is non-zero", notdos1x);
1556 goto out;
1559 for (i = 0; i < ARRAY_SIZE(floppy_defaults); i++) {
1560 if (floppy_defaults[i].nr_sectors == bd_sects) {
1561 fdefaults = &floppy_defaults[i];
1562 break;
1566 if (fdefaults == NULL) {
1567 if (!silent)
1568 fat_msg(sb, KERN_WARNING,
1569 "This looks like a DOS 1.x volume, but isn't a recognized floppy size (%llu sectors)",
1570 (u64)bd_sects);
1571 goto out;
1574 if (!silent)
1575 fat_msg(sb, KERN_INFO,
1576 "This looks like a DOS 1.x volume; assuming default BPB values");
1578 memset(bpb, 0, sizeof(*bpb));
1579 bpb->fat_sector_size = SECTOR_SIZE;
1580 bpb->fat_sec_per_clus = fdefaults->sec_per_clus;
1581 bpb->fat_reserved = 1;
1582 bpb->fat_fats = 2;
1583 bpb->fat_dir_entries = fdefaults->dir_entries;
1584 bpb->fat_sectors = fdefaults->nr_sectors;
1585 bpb->fat_fat_length = fdefaults->fat_length;
1587 error = 0;
1589 out:
1590 return error;
1594 * Read the super block of an MS-DOS FS.
1596 int fat_fill_super(struct super_block *sb, void *data, int silent, int isvfat,
1597 void (*setup)(struct super_block *))
1599 struct inode *root_inode = NULL, *fat_inode = NULL;
1600 struct inode *fsinfo_inode = NULL;
1601 struct buffer_head *bh;
1602 struct fat_bios_param_block bpb;
1603 struct msdos_sb_info *sbi;
1604 u16 logical_sector_size;
1605 u32 total_sectors, total_clusters, fat_clusters, rootdir_sectors;
1606 int debug;
1607 long error;
1608 char buf[50];
1611 * GFP_KERNEL is ok here, because while we do hold the
1612 * superblock lock, memory pressure can't call back into
1613 * the filesystem, since we're only just about to mount
1614 * it and have no inodes etc active!
1616 sbi = kzalloc(sizeof(struct msdos_sb_info), GFP_KERNEL);
1617 if (!sbi)
1618 return -ENOMEM;
1619 sb->s_fs_info = sbi;
1621 sb->s_flags |= MS_NODIRATIME;
1622 sb->s_magic = MSDOS_SUPER_MAGIC;
1623 sb->s_op = &fat_sops;
1624 sb->s_export_op = &fat_export_ops;
1625 mutex_init(&sbi->nfs_build_inode_lock);
1626 ratelimit_state_init(&sbi->ratelimit, DEFAULT_RATELIMIT_INTERVAL,
1627 DEFAULT_RATELIMIT_BURST);
1629 error = parse_options(sb, data, isvfat, silent, &debug, &sbi->options);
1630 if (error)
1631 goto out_fail;
1633 setup(sb); /* flavour-specific stuff that needs options */
1635 error = -EIO;
1636 sb_min_blocksize(sb, 512);
1637 bh = sb_bread(sb, 0);
1638 if (bh == NULL) {
1639 fat_msg(sb, KERN_ERR, "unable to read boot sector");
1640 goto out_fail;
1643 error = fat_read_bpb(sb, (struct fat_boot_sector *)bh->b_data, silent,
1644 &bpb);
1645 if (error == -EINVAL && sbi->options.dos1xfloppy)
1646 error = fat_read_static_bpb(sb,
1647 (struct fat_boot_sector *)bh->b_data, silent, &bpb);
1648 brelse(bh);
1650 if (error == -EINVAL)
1651 goto out_invalid;
1652 else if (error)
1653 goto out_fail;
1655 logical_sector_size = bpb.fat_sector_size;
1656 sbi->sec_per_clus = bpb.fat_sec_per_clus;
1658 error = -EIO;
1659 if (logical_sector_size < sb->s_blocksize) {
1660 fat_msg(sb, KERN_ERR, "logical sector size too small for device"
1661 " (logical sector size = %u)", logical_sector_size);
1662 goto out_fail;
1665 if (logical_sector_size > sb->s_blocksize) {
1666 struct buffer_head *bh_resize;
1668 if (!sb_set_blocksize(sb, logical_sector_size)) {
1669 fat_msg(sb, KERN_ERR, "unable to set blocksize %u",
1670 logical_sector_size);
1671 goto out_fail;
1674 /* Verify that the larger boot sector is fully readable */
1675 bh_resize = sb_bread(sb, 0);
1676 if (bh_resize == NULL) {
1677 fat_msg(sb, KERN_ERR, "unable to read boot sector"
1678 " (logical sector size = %lu)",
1679 sb->s_blocksize);
1680 goto out_fail;
1682 brelse(bh_resize);
1685 mutex_init(&sbi->s_lock);
1686 sbi->cluster_size = sb->s_blocksize * sbi->sec_per_clus;
1687 sbi->cluster_bits = ffs(sbi->cluster_size) - 1;
1688 sbi->fats = bpb.fat_fats;
1689 sbi->fat_bits = 0; /* Don't know yet */
1690 sbi->fat_start = bpb.fat_reserved;
1691 sbi->fat_length = bpb.fat_fat_length;
1692 sbi->root_cluster = 0;
1693 sbi->free_clusters = -1; /* Don't know yet */
1694 sbi->free_clus_valid = 0;
1695 sbi->prev_free = FAT_START_ENT;
1696 sb->s_maxbytes = 0xffffffff;
1698 if (!sbi->fat_length && bpb.fat32_length) {
1699 struct fat_boot_fsinfo *fsinfo;
1700 struct buffer_head *fsinfo_bh;
1702 /* Must be FAT32 */
1703 sbi->fat_bits = 32;
1704 sbi->fat_length = bpb.fat32_length;
1705 sbi->root_cluster = bpb.fat32_root_cluster;
1707 /* MC - if info_sector is 0, don't multiply by 0 */
1708 sbi->fsinfo_sector = bpb.fat32_info_sector;
1709 if (sbi->fsinfo_sector == 0)
1710 sbi->fsinfo_sector = 1;
1712 fsinfo_bh = sb_bread(sb, sbi->fsinfo_sector);
1713 if (fsinfo_bh == NULL) {
1714 fat_msg(sb, KERN_ERR, "bread failed, FSINFO block"
1715 " (sector = %lu)", sbi->fsinfo_sector);
1716 goto out_fail;
1719 fsinfo = (struct fat_boot_fsinfo *)fsinfo_bh->b_data;
1720 if (!IS_FSINFO(fsinfo)) {
1721 fat_msg(sb, KERN_WARNING, "Invalid FSINFO signature: "
1722 "0x%08x, 0x%08x (sector = %lu)",
1723 le32_to_cpu(fsinfo->signature1),
1724 le32_to_cpu(fsinfo->signature2),
1725 sbi->fsinfo_sector);
1726 } else {
1727 if (sbi->options.usefree)
1728 sbi->free_clus_valid = 1;
1729 sbi->free_clusters = le32_to_cpu(fsinfo->free_clusters);
1730 sbi->prev_free = le32_to_cpu(fsinfo->next_cluster);
1733 brelse(fsinfo_bh);
1736 /* interpret volume ID as a little endian 32 bit integer */
1737 if (sbi->fat_bits == 32)
1738 sbi->vol_id = bpb.fat32_vol_id;
1739 else /* fat 16 or 12 */
1740 sbi->vol_id = bpb.fat16_vol_id;
1742 sbi->dir_per_block = sb->s_blocksize / sizeof(struct msdos_dir_entry);
1743 sbi->dir_per_block_bits = ffs(sbi->dir_per_block) - 1;
1745 sbi->dir_start = sbi->fat_start + sbi->fats * sbi->fat_length;
1746 sbi->dir_entries = bpb.fat_dir_entries;
1747 if (sbi->dir_entries & (sbi->dir_per_block - 1)) {
1748 if (!silent)
1749 fat_msg(sb, KERN_ERR, "bogus number of directory entries"
1750 " (%u)", sbi->dir_entries);
1751 goto out_invalid;
1754 rootdir_sectors = sbi->dir_entries
1755 * sizeof(struct msdos_dir_entry) / sb->s_blocksize;
1756 sbi->data_start = sbi->dir_start + rootdir_sectors;
1757 total_sectors = bpb.fat_sectors;
1758 if (total_sectors == 0)
1759 total_sectors = bpb.fat_total_sect;
1761 total_clusters = (total_sectors - sbi->data_start) / sbi->sec_per_clus;
1763 if (sbi->fat_bits != 32)
1764 sbi->fat_bits = (total_clusters > MAX_FAT12) ? 16 : 12;
1766 /* some OSes set FAT_STATE_DIRTY and clean it on unmount. */
1767 if (sbi->fat_bits == 32)
1768 sbi->dirty = bpb.fat32_state & FAT_STATE_DIRTY;
1769 else /* fat 16 or 12 */
1770 sbi->dirty = bpb.fat16_state & FAT_STATE_DIRTY;
1772 /* check that FAT table does not overflow */
1773 fat_clusters = calc_fat_clusters(sb);
1774 total_clusters = min(total_clusters, fat_clusters - FAT_START_ENT);
1775 if (total_clusters > MAX_FAT(sb)) {
1776 if (!silent)
1777 fat_msg(sb, KERN_ERR, "count of clusters too big (%u)",
1778 total_clusters);
1779 goto out_invalid;
1782 sbi->max_cluster = total_clusters + FAT_START_ENT;
1783 /* check the free_clusters, it's not necessarily correct */
1784 if (sbi->free_clusters != -1 && sbi->free_clusters > total_clusters)
1785 sbi->free_clusters = -1;
1786 /* check the prev_free, it's not necessarily correct */
1787 sbi->prev_free %= sbi->max_cluster;
1788 if (sbi->prev_free < FAT_START_ENT)
1789 sbi->prev_free = FAT_START_ENT;
1791 /* set up enough so that it can read an inode */
1792 fat_hash_init(sb);
1793 dir_hash_init(sb);
1794 fat_ent_access_init(sb);
1797 * The low byte of FAT's first entry must have same value with
1798 * media-field. But in real world, too many devices is
1799 * writing wrong value. So, removed that validity check.
1801 * if (FAT_FIRST_ENT(sb, media) != first)
1804 error = -EINVAL;
1805 sprintf(buf, "cp%d", sbi->options.codepage);
1806 sbi->nls_disk = load_nls(buf);
1807 if (!sbi->nls_disk) {
1808 fat_msg(sb, KERN_ERR, "codepage %s not found", buf);
1809 goto out_fail;
1812 /* FIXME: utf8 is using iocharset for upper/lower conversion */
1813 if (sbi->options.isvfat) {
1814 sbi->nls_io = load_nls(sbi->options.iocharset);
1815 if (!sbi->nls_io) {
1816 fat_msg(sb, KERN_ERR, "IO charset %s not found",
1817 sbi->options.iocharset);
1818 goto out_fail;
1822 error = -ENOMEM;
1823 fat_inode = new_inode(sb);
1824 if (!fat_inode)
1825 goto out_fail;
1826 sbi->fat_inode = fat_inode;
1828 fsinfo_inode = new_inode(sb);
1829 if (!fsinfo_inode)
1830 goto out_fail;
1831 fsinfo_inode->i_ino = MSDOS_FSINFO_INO;
1832 sbi->fsinfo_inode = fsinfo_inode;
1833 insert_inode_hash(fsinfo_inode);
1835 root_inode = new_inode(sb);
1836 if (!root_inode)
1837 goto out_fail;
1838 root_inode->i_ino = MSDOS_ROOT_INO;
1839 root_inode->i_version = 1;
1840 error = fat_read_root(root_inode);
1841 if (error < 0) {
1842 iput(root_inode);
1843 goto out_fail;
1845 error = -ENOMEM;
1846 insert_inode_hash(root_inode);
1847 fat_attach(root_inode, 0);
1848 sb->s_root = d_make_root(root_inode);
1849 if (!sb->s_root) {
1850 fat_msg(sb, KERN_ERR, "get root inode failed");
1851 goto out_fail;
1854 if (sbi->options.discard) {
1855 struct request_queue *q = bdev_get_queue(sb->s_bdev);
1856 if (!blk_queue_discard(q))
1857 fat_msg(sb, KERN_WARNING,
1858 "mounting with \"discard\" option, but "
1859 "the device does not support discard");
1862 fat_set_state(sb, 1, 0);
1863 return 0;
1865 out_invalid:
1866 error = -EINVAL;
1867 if (!silent)
1868 fat_msg(sb, KERN_INFO, "Can't find a valid FAT filesystem");
1870 out_fail:
1871 if (fsinfo_inode)
1872 iput(fsinfo_inode);
1873 if (fat_inode)
1874 iput(fat_inode);
1875 unload_nls(sbi->nls_io);
1876 unload_nls(sbi->nls_disk);
1877 fat_reset_iocharset(&sbi->options);
1878 sb->s_fs_info = NULL;
1879 kfree(sbi);
1880 return error;
1883 EXPORT_SYMBOL_GPL(fat_fill_super);
1886 * helper function for fat_flush_inodes. This writes both the inode
1887 * and the file data blocks, waiting for in flight data blocks before
1888 * the start of the call. It does not wait for any io started
1889 * during the call
1891 static int writeback_inode(struct inode *inode)
1894 int ret;
1896 /* if we used wait=1, sync_inode_metadata waits for the io for the
1897 * inode to finish. So wait=0 is sent down to sync_inode_metadata
1898 * and filemap_fdatawrite is used for the data blocks
1900 ret = sync_inode_metadata(inode, 0);
1901 if (!ret)
1902 ret = filemap_fdatawrite(inode->i_mapping);
1903 return ret;
1907 * write data and metadata corresponding to i1 and i2. The io is
1908 * started but we do not wait for any of it to finish.
1910 * filemap_flush is used for the block device, so if there is a dirty
1911 * page for a block already in flight, we will not wait and start the
1912 * io over again
1914 int fat_flush_inodes(struct super_block *sb, struct inode *i1, struct inode *i2)
1916 int ret = 0;
1917 if (!MSDOS_SB(sb)->options.flush)
1918 return 0;
1919 if (i1)
1920 ret = writeback_inode(i1);
1921 if (!ret && i2)
1922 ret = writeback_inode(i2);
1923 if (!ret) {
1924 struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
1925 ret = filemap_flush(mapping);
1927 return ret;
1929 EXPORT_SYMBOL_GPL(fat_flush_inodes);
1931 static int __init init_fat_fs(void)
1933 int err;
1935 err = fat_cache_init();
1936 if (err)
1937 return err;
1939 err = fat_init_inodecache();
1940 if (err)
1941 goto failed;
1943 return 0;
1945 failed:
1946 fat_cache_destroy();
1947 return err;
1950 static void __exit exit_fat_fs(void)
1952 fat_cache_destroy();
1953 fat_destroy_inodecache();
1956 module_init(init_fat_fs)
1957 module_exit(exit_fat_fs)
1959 MODULE_LICENSE("GPL");